AMD: Speculation, Rumors, and Discussion (Archive)

[gamervivek]

Reddit users are claiming up to 30% better performance with the Vulkan patch on AMD cards.
AMD claims 27% on the RX480.

Given how their OpenGL performance was unexpectedly competitive, this is putting all GCN 1.1+ cards quite comfortably above their Maxwell counterparts.

AMD being competitive in OpenGL on Windows isn't really unexpected.

http://forums.anandtech.com/showthread.php?p=37391200#post37391200

 

[CSI PC]

Reddit users are claiming up to 30% better performance with the Vulkan patch on AMD cards.
AMD claims 27% on the RX480.

Given how their OpenGL performance was unexpectedly competitive, this is putting all GCN 1.1+ cards quite comfortably above their Maxwell counterparts.

Bear in mind that is reference cards for Nvidia, and the custom AIB for Maxwell 2 can be anywhere from 10% to 25% faster depending upon game and custom card.
As an example look at the games and compare the reference 980ti to the MSI 980ti Lightning that is a heck of a lot faster.
The review is for the 1070 but also includes reference and the MSI 980ti Lighting: http://www.pcgameshardware.de/Nvidi...als/GTX-1070-Benchmarks-Test-Preis-1196360/2/

Really needs both sets of measurements for perspective, as many buy custom cards rather than the reference Nvidia models for the better coolers and stock OC.
Appreciate the same can be said about AMD cards, but less so for stock OC custom models.

Curious to see some publications benchmark Doom with the Vulkan patch, just to get a feel how it behaves with both manufacturers, but looks like a good boost for AMD in general.
Cheers

 

[Kaotik]

And will the async compute be Pascal only like it is with the recent Tomb Raider patch? Meanwhile it appears that all GCN cards are likely to support it. And will there be as impressive a performance boost with Vulkan on Pascal as there is with GCN based cards?

Most likely it will be Pascal only, while Maxwell is technically capable of async, there's apparently no real way it could ever benefit from it due to it's limitations.
Pascals potential gains should probably be smaller than GCNs, since at least my understanding is that it isn't as flexible on the async front, which should make it harder to get gains from it

 

[DavidGraham]

The guys at DSOGaming are posting up to 60% uplift with 980Ti:

http://www.dsogaming.com/news/doom-...-vulkan-versus-opengl-performance-comparison/

 

[CSI PC]

The guys at DSOGaming are posting up to 60% uplift with 980Ti:

http://www.dsogaming.com/news/doom-...-vulkan-versus-opengl-performance-comparison/

Looks like it needs some careful analysis to see what is going on between those seeing performance boost and such as Guru3d, whether that be down to the map-scene or lighting/post processing/etc settings.

I remember the game having some heavy fps impact with certain shadow settings before this patch.
Cheers

 

[Grall]

AMD Blog Post about it:

Interesting that the performance benefit actually increases at 1440P... That seems rather counter-intuitive, unless 1080P is bottlenecked somewhere else than rendering. Where would that be, though?

 

[Silent_Buddha]

The guys at DSOGaming are posting up to 60% uplift with 980Ti:

http://www.dsogaming.com/news/doom-...-vulkan-versus-opengl-performance-comparison/

Interesting as the link in the Doom thread has the GTX 1080 getting far far less than that. If all we went by was those 2 things, it'd look like with Vulkan the 980Ti is a lot faster than the 1080. But I think I'll wait for actual reviews and real benchmarks done by quality sites before saying anything.

Regards,
SB

 

[SimBy]

Well you get almost 100% improvement in some scenes on AMD hardware. It's averages that matter though.

Guru3D did a quick run with RX480 and GTX1070.

http://www.guru3d.com/news-story/new-patch-brings-vulkan-support-to-doom.html

 

[Ike Turner]

Well you get almost 100% improvement in some scenes on AMD hardware. It's averages that matter though.

Guru3D did a quick run with RX480 and GTX1070.

http://www.guru3d.com/news-story/new-patch-brings-vulkan-support-to-doom.html

Yup I'm getting 100% improvements in select scenes on a Fury X + i7 6700K @ 1440P. I obviously wasn't really held back that much by my CPU and given the small perf improvements on Nvidia GPUs paired with similar beefy CPUs I'm guessing that Async Compute is really helping AMD's HW in those select scenes/encounters. Would be really interesting to see how things pan out once Async is patched to work on nVidia's GPUs.

Edit: Here's a GTX 1080 paired with an i7 6700K @ 1440P . OGL is actually faster than Vulkan

Kinda feels weird to see the tables turned now with DX12 and Vulkan where AMD is besting NVIDIA. Good riddance shitty APIs.

Blurry off-screen pic of the same scene, same settings; same res only difference is that it's my Fury X instead of a GTX 1080.

 

[lanek]

Looks like it needs some careful analysis to see what is going on between those seeing performance boost and such as Guru3d, whether that be down to the map-scene or lighting/post processing/etc settings.

I remember the game having some heavy fps impact with certain shadow settings before this patch.
Cheers

For be clear, DSO gaming point that they had abnormal performance on the 980TI on some specific outdoor zone ( not related to cpu or vram limitation ), and that by retesting them they have found this gain. the problem is they only have test ( yet ) the 980TI and outside thoses screenshoots, theres no real plot ( is this a peak fps, ( so increase of the max ), is their average Lets hope to have more data then.

 

[CSI PC]

Well you get almost 100% improvement in some scenes on AMD hardware. It's averages that matter though.

Guru3D did a quick run with RX480 and GTX1070.

http://www.guru3d.com/news-story/new-patch-brings-vulkan-support-to-doom.html

Also averages only tell part of the performance story if it does not include various map-scenes.
Look at Hitman, depending upon the episode or scene can mean the performance is faster for Nvidia or AMD with comparable tiers.

So analysis needs to work out the behaviour and then benchmark one or multiple scenes if they vary, same way PCGameshardware identified the DX12 memory protection issue in Hitman.
Some of the performance may come down to the shadow setting in Doom, which can have massive performance hit.
Cheers

 

[CSI PC]

For be clear, DSO gaming point that they had abnormal performance on the 980TI on some specific outdoor zone ( not related to cpu or vram limitation ), and that by retesting them they have found this gain. the problem is they only have test ( yet ) the 980TI and outside thoses screenshoots, theres no real plot ( is this a peak fps, ( so increase of the max ), is their average Lets hope to have more data then.

I am missing seeing that in the article.
They expressly mention the CPU though.

Did they update it somewhere else or comments?
They also included picture of an inside scene where the fps also increased from 148fps to 161fps.
BTW not saying this is normal behaviour, but then there are many variables that need to be accounted for and why a careful analysis is needed like I mentioned earlier.
Thanks

 

[lanek]

Also averages only tell part of the performance story if it does not include various map-scenes.
Look at Hitman, depending upon the episode or scene can mean the performance is faster for Nvidia or AMD with comparable tiers.

So analysis needs to work out the behaviour and then benchmark one or multiple scenes if they vary, same way PCGameshardware identified the DX12 memory protection issue in Hitman.
Some of the performance may come down to the shadow setting in Doom, which can have massive performance hit.
Cheers

Ofc, but you cant take only another zone in count too because it is there where you have a performance advantage too. DSO had point that the 980TI was seems to perform poorly on thoses outdoor scene, and seems point that with Vulkan they have the gain ? it was the same with the TitanX ? the same for 980?

Need more data yet.

I am missing seeing that in the article.
They expressly mention the CPU though.

Did they update it somewhere else or comments?
They also included picture of an inside scene where the fps also increased from 148fps to 161fps.
BTW not saying this is normal behaviour, but then there are many variables that need to be accounted for and why a careful analysis is needed like I mentioned earlier.
Thanks

Taken from the first line of their article:

As we wrote in our PC Performance Analysis, there were some scenes in DOOM during which our GPU usage was dropping for no apparent reasons.

“Now while DOOM runs exceptionally well on the PC, we did notice some optimization issues (that could be very well due to the OpenGL API). As we can see in the following screenshot, our GPU was underused for no apparent reason. In this scene – and other outdoor scenes – our GTX980Ti was not used to its fullest, even though we were not VRAM or CPU limited. Yes, we were still above 90-100fps, however it appears that the game does not fully utilize the GPU in specific scenes. It will be interesting to see whether these scenes run better with Vulkan.”

 

[CSI PC]

Ofc, but you cant take only another zone in count too because it is there where you have a performance advantage too. DSO had point that the 980TI was seems to perform poorly on thoses outdoor scene, and seems point that with Vulkan they have the gain ? it was the same with the TitanX ? the same for 980?

Need more data yet.

Where are they saying this is an anomaly of outside/abnormal performance?
Would help if you quote them as I cannot see what your mentioning on their analysis page, to me nothing is conclusive apart from the fps boost and it happening mostly outside for whatever reason.
It is possible Guru3d saw no benefits because of their own settings and map-scene, it goes both ways.
Why I mentioned further careful analysis is required taking the settings and map scene into consideration, and thread behaviour.
Thanks

 

[lanek]

Where are they saying this is an anomaly of outside?
Would help if you quote them as I cannot see what your mentioning on their analysis page, to me nothing is conclusive apart from the fps boost and it happening mostly outside for whatever reason.
Thanks

In general, when a GPU is called " unused " ( bottleneck ? ) for no apparent reason, thats an anomaly..

the problem yet is we have no data, no benchmark plots ( with min, max, average, lenght of the benchmarks, vram usage, ) different resolution, same plalce, other place, other gpu#s..

If the 980TI was act strangely there, should be the same with the TitanX ? ( if no vram limitation occur ).

I dont say that their numbers are wrong,. But we have no data yet lol from other reviewers, the only one are folks with 980TI on different forums who say they dont see any gain.

 

[CSI PC]

In general, when a GPU is called " unused " ( bottleneck ? ) for no apparent reason, thats a problem ..

Ah k,
you are taking the point that they noticed it not performing ideal before patch where they say:

Now while DOOM runs exceptionally well on the PC, we did notice some optimization issues (that could be very well due to the OpenGL API)...
Yes, we were still above 90-100fps, however it appears that the game does not fully utilize the GPU in specific scenes.
It will be interesting to see whether these scenes run better with Vulkan

So it looks like the Vulkan patch for them has improved areas where performance was weak before, explains why they also used those sections and maybe why Guru3d did not see any gains.
Going to be a headache to do decent analysis on, because the settings will also need to be considered and as I mentioned map zones along with the settings as they may interact with performance behaviour.
Thanks

 

[lanek]

Ah k,
you are taking the point that they noticed it not performing ideal before patch where they say:

So it looks like the Vulkan patch for them has improved areas where performance was weak before, explains why they also used those sections and maybe why Guru3d did not see any gains.
Going to be a headache to do decent analysis on, because the settings will also need to be considered and as I mentioned map zones along with the settings.
Thanks

Sorry i was not much clear ( my bad english sometimes dont help )..

 

[CSI PC]

Sorry i was not much clear ( my bad english sometimes dont help )..

No need for sorry we got there
And the point-context you raise is important, also it became confusing as we both posted a response to each other at the same time
Cheers

 

[Ike Turner]

OGL vs Vulkan W/ Async Off vs Vulkan w/ Async On (1080P+ Ultra settings on a RX480 4GB)

Big FPS difference in some cases (mainly when no shooting)

 

[pharma]

RX 480 8GB CrossFire Review

Using 60% more power than GTX 1080, the performance benefit doesn't seem to be there in regards to AMD Radeon RX 480 8GB CrossFire. 60% more system power required for performance that may at times equal the GeForce GTX 1080, but other times may not, is not a good value at all.

Yes, AMD Radeon RX 480 8GB CrossFire costs less than GeForce GTX 1080, a whole lot less, and compares closer to GeForce GTX 1070 on price. AMD Radeon RX 480 8GB CrossFire may be a good value compared to GeForce GTX 1070, offering more performance, for a somewhat equivalent price. That is where it seems to fit best. Compared to GeForce GTX 1080 though, the GTX 1080 is more consistent in every way, performance and frametime offering a smoother experience, literally.

http://www.hardocp.com/article/2016/07/11/amd_radeon_rx_480_8gb_crossfire_review/1